Detecting Adverse Events Following Acute Care Encounters among Veterans Prescribed Non-Steroidal Anti-Inflammatory Drugs.

OBJECTIVES: To model the occurrence of adverse events (AEs) among Veterans prescribed nonsteroidal anti-inflammatory drugs (NSAIDs) during emergency department and urgent care encounters within the Veterans Health Administration (VHA) care setting. NSAIDs can increase the risk for gastrointestinal, renal, and cardiovascular AEs, particularly in older adults, patients with chronic kidney disease (CKD), or patients with polypharmacy. METHODS: We applied eXtreme Gradient Boosting (XGBoost) modeling and least absolute shrinkage and selection operator (LASSO) methods along with logistic regression to model the occurrence of AEs in a cohort of Veterans prescribed NSAIDs seen between January 1, 2017, and December 31, 2023. The outcome was defined as a composite of acute coronary syndrome, acute kidney injury, allergic reactions, gastrointestinal bleeding, or gastroesophageal reflux disease (GERD) within 2 to 30 days following the encounter. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) values and calibration plots. Observed-to-expected (OE) ratios for AEs were calculated. RESULTS: We identified 1,112,819 acute care encounters for 736,677 unique patients and a total of 50,278 (4.5%) AEs. The XGBoost and LASSO regression models performed well across time, with AUC values ranging from 0.77 to 0.79. GERD and a diagnosis of CKD stage 4 or 5/dialysis were identified as having a harmful association with AEs. The OE ratios varied by VHA care site. CONCLUSIONS: The risk adjustment models had good discrimination performance across time and allowed us to identify significant risk-adjusted site variation. The models may be useful in clinical decision support and population health interventions with NSAIDs.